23 ms·
Pretty neat, though it looks like the sentiment analysis could use some work. There are quite a lot of negative ones on the positive side. EDIT: Might be usefu
by jstanley 14y ago
Pretty neat, though it looks like the sentiment analysis could use some work. There are quite a lot of negative ones on the positive side.
EDIT: Might be useful to expand this into a generic service where people can enter their own search terms.
- locopati 14y agoOr allow page viewers to move something from one column to the other and have the back-end receive notification so that it modifies its sentiment algorithm automagically.
- ScottBurson 14y agoI see some positive ones on the negative side too: "Batman no!! You can't lose your faith in humanity!" "You must not lose faith in humanity. [...] - Gandhi" I don't know how the sentiment analysis is being done (naive Bayes?), but perhaps it could be tweaked to handle this case.
- Fishkins 14y agoEven a tweet about this website itself is categorized as negative: https://twitter.com/LeviDayley/status/313711985437073409 https://twitter.com/LeviDayley/status/313711985437073409
- kafkaesque 14y agoHere is a stab in the dark: These types of declaratives have two negatives side by side (e.g., "not lose", "can't lose"), making them into a positive. They're somewhat similar to rhetorical questions which have assertions embedded in them in the form of a negative question. From the Wikipedia article, "What have the Romans ever done for us?..." (What...ever...?) "...should be read as The Romans have never done anything for us"... (where a negative is surrounded by a verb). http://en.wikipedia.org/wiki/Rhetorical_question#Negative_assertions http://en.wikipedia.org/wiki/Rhetorical_question#Negative_as... What do people think of this? I'm not entirely familiar with naive Bayes classifier.
- anigbrowl 14y agoThey'll be cancelled out by false positives, eg 'I try to maintain my my faith in humanity. Sometimes I succeed.'
- imkevinxu 14y agoHi! Creator here, yes the sentiment analysis could definitely use more work but we've found our naiive approach to be about 95% accurate which was good enough for our MVP. Definite idea, some friends have mentioned it might be useful for brands
- fatjokes 14y agoMight I ask what exactly is the naive approach? Is it simply searching for keywords like "lose", "lost", etc?
- callmecosmas 14y agoPretty much: https://github.com/imkevinxu/faithinhumanity/blob/master/faithinhumanity_app/management/commands/tweet_scraper.py https://github.com/imkevinxu/faithinhumanity/blob/master/fai...